Epileptic Seizure Classification with Multivariate Empirical Mode Decomposition and Hilbert Vibration Decomposition

EEG signals are frequently used to record seizures of epilepsy. However, observation of these seizures is difficult and time-consuming. Fourier-based approaches are not suitable for the nonlinear and nonstationary nature of EEG. For this reason, empirical methods such as multivariate empirical mode decomposition (MEMD) are used in the analysis of epileptic EEG signals. This study compares MEMD with Hilbert vibration decomposition method (HVD) in the classification of EEG signals in terms of epileptic status. Frequency and entropy-based attributes were obtained from subcomponents obtained by these two methods and classification was made by convolutional neural network, random forest and support vector machine classifiers. As the result of the study, the HVD method showed higher performance than the MEMD method and reached 100% classification accuracy.